[dm-devel] [PATCH v2 3/4] nvme: introduce FAILUP handling for REQ_FAILFAST_TRANSPORT

[Hannes Reinecke]

On 4/16/21 5:03 PM, Mike Snitzer wrote:
> On Fri, Apr 16 2021 at 10:07am -0400,
> Hannes Reinecke <hare at suse.de> wrote:
> 
>> On 4/16/21 1:15 AM, Mike Snitzer wrote:
>>> If REQ_FAILFAST_TRANSPORT is set it means the driver should not retry
>>> IO that completed with transport errors. REQ_FAILFAST_TRANSPORT is
>>> set by multipathing software (e.g. dm-multipath) before it issues IO.
>>>
>>> Update NVMe to allow failover of requests marked with either
>>> REQ_NVME_MPATH or REQ_FAILFAST_TRANSPORT. This allows such requests
>>> to be given a disposition of either FAILOVER or FAILUP respectively.
>>> FAILUP handling ensures a retryable error is returned up from NVMe.
>>>
>>> Introduce nvme_failup_req() for use in nvme_complete_rq() if
>>> nvme_decide_disposition() returns FAILUP. nvme_failup_req() ensures
>>> the request is completed with a retryable IO error when appropriate.
>>> __nvme_end_req() was factored out for use by both nvme_end_req() and
>>> nvme_failup_req().
>>>
>>> Signed-off-by: Mike Snitzer <snitzer at redhat.com>
>>> ---
>>>  drivers/nvme/host/core.c | 31 ++++++++++++++++++++++++++-----
>>>  1 file changed, 26 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
>>> index 4134cf3c7e48..10375197dd53 100644
>>> --- a/drivers/nvme/host/core.c
>>> +++ b/drivers/nvme/host/core.c
>>> @@ -299,6 +299,7 @@ enum nvme_disposition {
>>>  	COMPLETE,
>>>  	RETRY,
>>>  	FAILOVER,
>>> +	FAILUP,
>>>  };
>>>  
>>>  static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
>>> @@ -318,10 +319,11 @@ static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
>>>  	    nvme_req(req)->retries >= nvme_max_retries)
>>>  		return COMPLETE;
>>>  
>>> -	if (req->cmd_flags & REQ_NVME_MPATH) {
>>> +	if (req->cmd_flags & (REQ_NVME_MPATH | REQ_FAILFAST_TRANSPORT)) {
>>>  		if (nvme_is_path_error(nvme_req(req)->status) ||
>>>  		    blk_queue_dying(req->q))
>>> -			return FAILOVER;
>>> +			return (req->cmd_flags & REQ_NVME_MPATH) ?
>>> +				FAILOVER : FAILUP;
>>>  	} else {
>>>  		if (blk_queue_dying(req->q))
>>>  			return COMPLETE;
>>> @@ -330,10 +332,8 @@ static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
>>>  	return RETRY;
>>>  }
>>>  
>>> -static inline void nvme_end_req(struct request *req)
>>> +static inline void __nvme_end_req(struct request *req, blk_status_t status)
>>>  {
>>> -	blk_status_t status = nvme_error_status(nvme_req(req)->status);
>>> -
>>>  	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
>>>  	    req_op(req) == REQ_OP_ZONE_APPEND)
>>>  		req->__sector = nvme_lba_to_sect(req->q->queuedata,
>>> @@ -343,6 +343,24 @@ static inline void nvme_end_req(struct request *req)
>>>  	blk_mq_end_request(req, status);
>>>  }
>>>  
>>> +static inline void nvme_end_req(struct request *req)
>>> +{
>>> +	__nvme_end_req(req, nvme_error_status(nvme_req(req)->status));
>>> +}
>>> +
>>> +static void nvme_failup_req(struct request *req)
>>> +{
>>> +	blk_status_t status = nvme_error_status(nvme_req(req)->status);
>>> +
>>> +	if (WARN_ON_ONCE(!blk_path_error(status))) {
>>> +		pr_debug("Request meant for failover but blk_status_t (errno=%d) was not retryable.\n",
>>> +			 blk_status_to_errno(status));
>>> +		status = BLK_STS_IOERR;
>>> +	}
>>> +
>>> +	__nvme_end_req(req, status);
>>> +}
>>> +
>>>  void nvme_complete_rq(struct request *req)
>>>  {
>>>  	trace_nvme_complete_rq(req);
>>> @@ -361,6 +379,9 @@ void nvme_complete_rq(struct request *req)
>>>  	case FAILOVER:
>>>  		nvme_failover_req(req);
>>>  		return;
>>> +	case FAILUP:
>>> +		nvme_failup_req(req);
>>> +		return;
>>>  	}
>>>  }
>>>  EXPORT_SYMBOL_GPL(nvme_complete_rq);
>>>
>>
>> Hmm. Quite convoluted, methinks.
> 
> Maybe you didn't read the header or patch?
> 
> I'm cool with critical review when it is clear the reviewer fully
> understands the patch but... ;)
> 
>> Shouldn't this achieve the same thing?
>>
>> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
>> index e89ec2522ca6..8c36a2196b66 100644
>> --- a/drivers/nvme/host/core.c
>> +++ b/drivers/nvme/host/core.c
>> @@ -303,8 +303,10 @@ static inline enum nvme_disposition
>> nvme_decide_disposition(struct request *req)
>>         if (likely(nvme_req(req)->status == 0))
>>                 return COMPLETE;
>>
>> -       if (blk_noretry_request(req) ||
>> -           (nvme_req(req)->status & NVME_SC_DNR) ||
>> +       if (blk_noretry_request(req))
>> +               nvme_req(req)->status |= NVME_SC_DNR;
>> +
>> +       if ((nvme_req(req)->status & NVME_SC_DNR) ||
>>             nvme_req(req)->retries >= nvme_max_retries)
>>                 return COMPLETE;
> 
> Definitely won't achieve the same. And especially not with patch 1/4
> ("nvme: return BLK_STS_DO_NOT_RETRY if the DNR bit is set") that you
> gave your Reviewed-by to earlier.
> 
Ah. Right. Sorry.

> Instead of "FAILUP", I thought about using "FAILUP_AND_OVER" to convey
> that this is a variant of failover.  Meaning it takes the same patch as
> nvme "FAILOVER" until the very end; where it does REQ_FAILFAST_TRANSPORT
> specific work detailed in nvme_failup_req().
> 
All very intricate; will need to check the patches in their combined
version.
Not deliberately stalling, mind you, just wanting to figure out what the
net result will be.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		        Kernel Storage Architect
hare at suse.de			               +49 911 74053 688
SUSE Software Solutions Germany GmbH, 90409 Nürnberg
GF: F. Imendörffer, HRB 36809 (AG Nürnberg)